Imagine you’ve built a startup from nothing to hundreds of thousands of dollars in recurring revenue. You’re feeling proud but a little burned out – maybe anxious to start something new. You decide to sell up, take a year out to rest and recharge, and then apply everything you’ve learned to a new venture.
But what if the market isn’t ready? What if you can’t achieve the price you’re looking for because buyer appetites have changed? Do you give up and slog on until you crash? Or would you rather diagnose and solve the problem in your business early and then list with better odds of success?
Determining whether your startup will sell is no easy feat. Our curation team can advise you, but we won’t always get it right. Countless factors come into play, many of which – like market conditions and your business performance – are constantly in flux. Could artificial intelligence be the answer?
Whether you’re embracing AI or fearing a robot uprising, you can’t deny the technology is excellent at spotting patterns that humans miss. In some studies, it’s more accurate than human analysts – and faster, too. Apply that power to your acquisition, and you reveal powerful insights.
Never one to miss a chance at improving your acquisition experience, we recently started training AI on past acquisition data. The results are promising so far, and I’d love to share how we’re building machine-learning models for everything from calculating your odds of selling to due diligence checks.
Why Use AI as a Diagnostic Tool in Acquisitions?
Our curation team spends all day determining which startups to add to the marketplace. Strict acceptance criteria mean only around 45 percent (or less) of startups make the grade. This isn’t so much a judgment on your business but a reflection of your likelihood of getting acquired. We know what buyers want, and we’d be doing you a disservice by listing a company that wouldn’t sell.
But this is where the challenge arises. Terms like “likelihood” and “probably” suggest statistical and analytical judgments that are only as reliable as the data underpinning them. While our curation team excels at identifying the fundamentals of a good business (criteria that most would agree make startups attractive acquisition targets), these evaluations don’t always reflect what unfolds once a listing goes live.
Some startups don’t sell as quickly as we believed they might, for example. Others that barely scrape past our acceptance criteria get acquired faster than we expected based on current trends. We’re already using past acquisition data to predict future success, but until now, we’ve been unable to do so efficiently, and at scale, a data problem many companies solve today with artificial intelligence.
We believe by training a machine learning model on the correct data, we’ll be able to calculate the probability of a startup selling while pinpointing why startups with potential don’t sell. You can then use this data before you list to address problems that might’ve slowed or ruined your acquisition. Otherwise, you risk entering the market unprepared for its reaction and may lose the advantage.
How We’re Training AI to Evaluate Startups
The first step in training artificial intelligence lies in data collection and preparation. AI is only as good as the data you train it on. Feed it everything, and you risk skewing output in a certain direction. Put another way, AI tends to amplify outliers and hidden prejudices. Acquisitions at 100x multiples, for example, or pre-revenue startups that repel most buyers from making offers.
In addition to acquisition data, we also collect market trends, economic data, and other relevant information that influences acquisitions. Macroeconomic factors become increasingly important as you develop and test a machine learning model. For example, you might need to test different data timeframes and adjust for changes like post-Covid digitalization or other temporary trends.
Once you’ve trained AI to spot patterns and trends in your dataset, you must validate its ability to “learn” from this data to predict outputs. Retaining some of the training data helped us here, otherwise we’d have needed to generate new data for the testing phase, which would’ve slowed development.
But there’s always a limit to how fast you can train an AI model. Especially when using data from such a complex industry as M&A where thousands of data points come into play. While providers like AWS and Google Cloud make training your own AI models accessible for almost everyone, costs quickly skyrocket the more data you feed the model. In other words, you need to balance getting your model ready and the high cost of processing vast volumes of data.
The testing phase is where you refine and tweak the model so it does a good enough job most of the time that you can trust in a production environment. No AI model is perfect. The goal is to get it good enough that its predictions are helpful to the people using it within the limits of the technology. That bears repeating: the goal isn’t to replace a professional’s opinion but to augment it using technology.
How You Could Benefit From the Technology
Using AI to predict the sellability of a business could increase the acquisition success rate of companies sold on the marketplace. In short, we can better assess whether your startup will sell, which augments our curation team’s knowledge and expertise, minimizing the rejection of potentially sellable startups. As a result, the acquisition process may seem less stressful and uncertain and more rewarding.
If your business scores low, the tool could identify shortcomings to address before listing. For example, maybe you need to update your asking price or business description to increase its appeal. How this works in practice is still to be decided. You might be able to see your score and get instant guidance. Or, it’ll play a background role within the curation team, where your score will join the hundreds of other factors the team considers when curating listings. Either way, you’ll always get the best guidance.
Ultimately, we hope the tool will increase the scope and sellability of startups on the marketplace. Attracting serious buyers ensures you get offers. And by populating the marketplace with high-quality listings that buyers want to acquire, we’ll attract even more buyers, and the circle continues, increasing your pool of potential candidates and reaffirming your chances of a successful exit.
As time passes, we expect the machine learning models we train to improve. The recently introduced buyer and seller feedback feature, for example, will enable us to train the models on emerging data and trends that boost its accuracy. Combining machine learning predictions with real-time market feedback will enhance our understanding of what makes a startup desirable. If a startup scores low but still gets acquired, we can refine the tool using the new data.
The Ethics of Using AI to Predict Acquisition Success
Outliers, as mentioned before, can and do sell on Acquire.com. Just because they represent the 1-5 percent of cases doesn’t mean we should deny their chance to sell. As such, any tool we build using artificial intelligence will not replace our curation team’s judgment of the market. If we think a low-scoring startup will sell, we’ll list it. Unlike the M&A team, the tool won’t be on personal terms with buyers, which is often how challenging startups sell – through our advisors’ vast buyer network.
We also plan to be as transparent as possible with how the tool works. AI is still a black box – no one is sure how a trained algorithm makes decisions. While researchers and engineers continue to work on AI transparency, we’ll continue helping you understand your odds of selling and how it influences our decision-making. Our curation team already does this when giving feedback following submissions. Almost all listings, even those we accept, receive optimization tips.
You’ll also learn more about our work with AI, such as the data we use to train our models, if we feel it would clarify the guidance we give you. AI might enhance our ability to predict a startup’s chances of selling, but it doesn’t tell the entire story. We must continue to train the model with new data, perhaps even baking it into a virtual advisor you can interact with for basic M&A guidance, allowing your actual M&A advisor to work on larger problems.
The applications for artificial intelligence in M&A are staggering. Intelligent matching of buyers to startups. Sellability scores. Due diligence management. Diagnostic insights. Personalized browsing experiences. While you’ll always receive the help of an expert throughout your exit, expect greater automation and deeper insights that help you (and us) make better decisions. All of which guides you to achieving a life-changing acquisition.
The content on this site is not intended to provide legal, financial or M&A advice. It is for information purposes only, and any links provided are for your convenience. Please seek the services of an M&A professional before entering into any M&A transaction. It is not Acquire’s intention to solicit or interfere with any established relationship you may have with any M&A professional.
The post What If You Knew the Odds of Your Startup Getting Acquired? appeared first on Acquire.com Blog.